11h30 - 11h55
Prediction of chronic damage in Systemic Lupus Erythematosus by using machine-learning models
The increased survival in Systemic Lupus Erythematosus (SLE) patients implies the development of chronic damage, occurring in up to 50% of cases. Its prevention is a major goal in the SLE management. Our aim is to predict the development of chronic damage, assessed by the SLICC/ACR Damage Index (SDI), by using neural networks. In particular, we applied Recurrent Neural Networks (RNNs) as a machine-learning model to deal with the sequential nature of the medical records (the visits of the patients through time). Two groups of patients were analyzed: patients with SDI=0 at the baseline, developing damage during the follow-up, and patients without damage (SDI=0). We used the data, comprising laboratory and clinical items for 138 patients, from the Sapienza Lupus Cohort, to train and test the model. The trained model achieved an AUC value of 0.77 on an 8-fold test showing the validity of the proposed machine learning-based approach to predict damage development.
11h55 - 12h20
On Support Vector Machines for Functional Data
Support Vector Machine (SVM) classifiers, obtained by solving a concave quadratic maximization problem with linear constraints, are a powerful tool in supervised classification. In this talk we address the application of SVM to classify functional data, focusing in two challenging issues: i) how to tune functional parameters, and ii) how to perform variable selection. Both problems are expressed as global optimization problems, for which procedures which exploit the problem structure are proposed. Computational results demonstrate the usefulness of our approach
12h20 - 12h45
Supervised and unsupervised learners to detect postural disorders
We present an applied project developed in collaboration with the Anatomics, Histology, Legal Medicine and Orthopedics Department of Sapienza University of Rome. The aim of the project is developing a new approach to postural disorders detection based on a combination of supervised and unsupervised learning techniques. We use data obtained by the Formetric machine which reports spinal column and posture measures.
Although the possibility to record large amounts of data from patients in the orthopedics field, datasets are still narrow and a method to automatically detect postural disorders in patients is missing. Furthermore, the influence of pathological conditions is unclear.
We propose a new approach based on classification and clustering techniques, in order to look for the variables that can support an orthopedist in the diagnostic process and define a learner to determine whether a patient is affected by a postural disorder is present or not.